AI

The Algorithmic Accountability Act is back—here’s what’s in it

The bill, originally proposed in 2019, could be the first federal AI regulation in the US.
article cover

Dkfielding/Getty Images

· 3 min read

Stay up to date on emerging tech

Drones, automation, AI, and more. The technologies that will shape the future of business, all in one newsletter.

The Algorithmic Accountability Act is back—and better than ever, according to the US lawmakers behind it.

Sen. Ron Wyden, Sen. Cory Booker, and Rep. Yvette Clarke first introduced the bill in 2019, as a way to hold companies accountable for “flawed computer algorithms that result in inaccurate, unfair, biased, or discriminatory decisions impacting Americans.” Nearly three years later, they’ve introduced a new version, after consulting experts and advocacy groups about improvements.

If passed, it would be among the first federal regulations for algorithms in the US, though we’ve seen some oversight at state and city levels around specific AI use cases. The EU is also considering a comprehensive Artificial Intelligence Act.

What’s inside

The bill would require companies to conduct impact assessments for algorithmic decision-making systems. Those assessments would include, among other measures, a section on the manual process the algorithmic system is replacing and why it’s necessary, a description of the privacy risks and potential downstream negative impacts on consumers’ safety and security, an evaluation to see if the system treats consumers differently according to race, sex, gender, age, disability, religion and more.

The FTC’s homework: The Act would also require that the Federal Trade Commission create regulations for the assessments, collect documentation from companies, and publish an aggregate report on trends every year. Plus, the FTC must establish a public repository of companies’ systems, including data sources, metrics, and a way to contest the algorithms’ decisions.

  • The bill would also add 50 to 75 FTC staff members and establish a Bureau of Technology to enforce the new legislation.

Who’s affected: Any company that deploys an automated system to make “critical decisions”—e.g., affecting a consumer’s access to or cost of education, employment, utilities, family planning, financial services, healthcare, housing, legal services, and more.

  • For the bill to apply, a company needs to be using identifying info (think: name, address, phone number, biometrics, etc.), have 1+ million consumers, and generate either $50+ million in average annual gross receipts or have $250+ million in annual equity value.

Big picture: We live in a world where algorithms can help decide who gets into college, who is offered a job interview, who receives a home loan, and more—and over the past year, controversy over AI has come to a head over the current lack of oversight, regulation, and formal auditing processes for algorithms.

Some experts believe 2022 is the year we’ll see concrete AI policy take shape. It’s unclear whether the Algorithmic Accountability Act of 2022 will receive enough votes to move forward, but it does have more co-sponsors and industry support than its former iteration.

Stay up to date on emerging tech

Drones, automation, AI, and more. The technologies that will shape the future of business, all in one newsletter.